Prepare the source

Identify which sounds and off-screen voices a viewer needs to understand the media. A transcript of dialogue alone may be incomplete.

Check speaker identification, relevant music and sound descriptions alongside the words. Before processing a full recording, test a passage that includes the difficult conditions you expect in the rest of the material. A clean introduction may not reveal problems with quieter answers, room noise or changes of language later in the file.

Settings for this workflow

Set the audio language when you know it, and begin with Tiny to confirm the device can run the model. Compare another available model on the same passage if the first draft needs too much correction. Keep the task on original-language transcription unless an English translation is specifically the goal.

DecisionPractical choice
Review priorityCheck speaker identification, relevant music and sound descriptions alongside the words.
Recommended exportVTT for a web player or the format required by the delivery platform.
Boundary to rememberA passing character-speed check does not establish accessibility conformance. Human review of content and playback is still required.

Review against the recording

Add the missing sound information manually. Test synchronization, line breaks and whether text obscures useful visual content.

Use the timestamp beside a segment to return to its audio. Correct the draft before copying a passage into another document. If a phrase remains unclear, leave an explicit marker in your working copy so the uncertainty survives later editing and sharing.

Illustrative handoff

Imagine a reviewer receiving the draft without having attended the session. They need the source recording identifier, the chosen transcription convention, a clear distinction between spoken words and notes, and a way to find unresolved passages. Make those choices visible in the exported document instead of assuming the next person will infer them.

VTT for a web player or the format required by the delivery platform. Keep the source audio under the access rules for this project. The app stores a local copy when browser storage permits; that convenience is not a substitute for an approved archive or backup.

Know the limit

A passing character-speed check does not establish accessibility conformance. Human review of content and playback is still required. If the work requires a managed transcription service, independent checking or a specialized record format, use a workflow that provides it. The useful result here is a transparent, reviewable draft.

Sources

  1. Whisper paper, appendix tables and normalization method · OpenAI / Radford et al. · Checked 2026-10-30
  2. Browser decoding and resampling · MDN · Checked 2026-10-30
  3. W3C prerecorded captions guidance · W3C WAI · Checked 2026-10-30
  4. Netflix English subtitle style guide · Netflix · Checked 2026-10-30